EAC-Net Charge Density Dataset: 150k Sampled Grids from Materials Project CHGCAR Subset for Equivariant Learning

This dataset contains processed charge-density samples derived from Materials Project (MP) CHGCAR files, used in training the EAC-mp large model. The raw CHGCAR files were obtained via the MP API, and from each frame 5,000 grid points were randomly sampled. In total, the dataset comprises 149,196 frames.

The dataset provides:

1. Per-frame metadata (task ID and structure information);

2. Sampled grid point coordinates and corresponding charge-density values;

3. MD5 checksums for data integrity.

Provenance & source: Data derived from Materials Project CHGCAR files (accessed August 2025). Users should cite both the Materials Project and the EAC-Net paper (arXiv:2508.04052) when using this dataset.

Usage notes: This is a processed and downsampled dataset, not a full CHGCAR archive. For access to the original CHGCAR files, please obtain them directly from the Materials Project in accordance with MP’s terms of use.